You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
pstack: Hillclimb opens its log "via show-me-your-work" but defines its own file name and columns #1297
Hillclimb says to open its decision log through show-me-your-work, then defines a different file and a different set of columns.
show-me-your-work asks callers not to do that.
Open the decision log via the show-me-your-work skill. A decision.tsv, one row per attempt: id, hypothesis, change, before, after, delta, tests, verdict (kept or reverted), note.
The Reply line asks for "the decision.tsv path" (hillclimb.md:21).
show-me-your-work:
Has one canonical format with six columns: ts, phase, decision, why, evidence, result (SKILL.md:9-22).
Its helper scripts/log.sh takes exactly those fields and exits 1 on any other argument count, so it cannot write a Hillclimb row.
Keeps the log at decisions.tsv or .audit/<task-slug>.tsv (SKILL.md:46).
Says: "Other skills route their audit trail here instead of inventing one. Reference it by name and let it own the format. Don't restate the columns." (SKILL.md:80-82)
In practice an agent running Hillclimb has to pick one of the two formats.
Either log.sh cannot be used, or the per-attempt fields Hillclimb reads before each attempt (before, after, delta, verdict) end up packed into decision, evidence and result.
Possible fixes:
Have Hillclimb use the show-me-your-work format and say how an attempt maps onto it.
For example: phase is the attempt id, decision is the change, why is the hypothesis, evidence is the commit and the before and after numbers, and result is kept or reverted with the delta.
Keep Hillclimb's nine-column table as its own experiment log, drop "via the show-me-your-work skill" from step 3, and log only decision points to show-me-your-work as usual.
Activity